Open Stack

On Fri, Feb 07, 2014 at 03:58:28PM +0100, Salvatore Orlando wrote:
> I just spoke with Eugene over IRC and the problem is that the test attempts
> to connect to an instance over a private network, which cannot work in
> isolated mode.
> This patch will skip the load balancing scenario:
> https://review.openstack.org/#/c/71898/
> 
> So thankfully, there is no new neutron issue.

So I'll ping him on IRC but I'd like to see a path for making that test work in
the gate otherwise I think we should just pull it. Otherwise it'll never be run
with other proposed commits and will probably bitrot fairly quickly.

> 
> But still, we should probably make gate tests symmetric again.
> 
> Salvatore
> 
> 
> On 7 February 2014 15:05, Salvatore Orlando <sorlando at nicira.com> wrote:
> 
> > Hi,
> >
> > By merging patch https://review.openstack.org/#/c/53459/, full tenant
> > isolation has been turned on for neutron API tests.
> > We have executed a few experimental checks before merging and we found out
> > that the isolated jobs were always passing.
> >
> > However, the gating jobs both in tempest and nova do not turn on tenant
> > isolation for neutron.
> > As a result we are unable to know how a new test or nova patch might
> > affects the neutron isolated jobs, on which neutron is gating.
> >
> > In particular, the patch https://review.openstack.org/#/c/58697/introduces a load balancing scenario which fails the isolated jobs (only
> > them) most of the time. The scenario itself seems fairly well designed and
> > therefore it might have uncovered a neutron issue.
> >
> > The problem is that now the neutron gate is seeing a high failure rate,
> > whereas if we turned tenant isolation on for gating jobs in tempest and
> > nova we would have found out about this issue before merging the load
> > balancing scenario. I think therefore it might be a good idea to:
> >
> > - switch neutron jobs to isolated in nova and tempest

So the flag to make tenant isolation was always a temporary thing just for
neutron so we could work out stabilizing things. It's been on by default for
all the other jobs for as long as I can remember. If you think that we've
reached the point were we can greenlight it everywhere, I'll push out the infra
changes to do that.

> > - do not executed anymore non-isolated jobs for neutron. I think this was
> > always the plan anyway.
> >

Yeah, that makes sense, but I still think we should keep 4 tempest jobs running
for the neutron queue. Do you have any ideas for other configs or should we just
duplicate the existing 2? (postgres and mysql)


-Matt